Information Technology Reference
In-Depth Information
and can reflect the status of the welding process from the weld pool. In addition, it
includes sufficient information about weld pool which is useful for process modeling
and control. Many researches were conducted to apply vision method in welding
process monitoring.
In Ref. [8]-[16], two dimensional measurements of weld pool were obtained, and
model establishments, penetration control were conducted. By sensing two
dimensional weld pool images, certain characteristic parameters and penetration
control can be achieved. Furthermore, three dimensional weld pool shapes presents
more sufficient information for determining the state of weld pool, analyzing pool
defect, conducting penetration control, and studying pool physical process.
The weld pools were monitored and reconstructed in three dimensional [17]-[23].
Besides, multi-variable process model were established, and advanced control were
conducted to obtain desired penetration control. Recently, a novel three-dimensional
monitoring system was developed and the observed three-dimensional weld pool was
characterized by its width, length and convexity [24]. An optimal linear model was
also obtained in steady state.
In this paper, an advanced three dimensional monitoring system developed at the
University of Kentucky [24] was utilized to sense the three-dimensional weld pool
surface in GTAW process. Then, novel characteristic parameters were defined and
welding experiments were conducted to obtain the process data. At last, a neural
network model was established to predict the back side parameters precisely. Based
on the proposed intelligent sensing model, the 3D monitoring system can
functionalize as skilled welders to observe the weld pool and predict the penetration
status.
2
System Structure and Image Processing
System structure for GTAW control system is shown in Fig.1 [24]. Based on the three
dimensional sensing system and weld pool reconstruction, the weld pool shape and
relative characteristic parameters were obtained. Then, the process soft-sensing model
was established after the process dynamic characteristic was observed. This model
can be used to estimate the penetration status online and provide necessary feedback
for the advanced controller.
The experiment system is shown in Fig.2 [24]. Gas tungsten arc welding without
filler for pipe welding was exploited in the system. A 20mw illumination laser at a
wavelength of 685nm with variable focus was used to project a 19-by-19 dot matrix
structured light pattern (Model SNF-519X (0.77)-685-20) on the weld pool surface in
the pipe. An imaging plane was fastened coaxially to the laser to intercept the
reflected laser rays. The intensity of laser almost remained as the same when the rays
intercepted by the imaging plane while the intensity of the arc light decays
significantly by distance. Then, the image was captured by the camera in real time at
certain frequency. Furthermore, the geometry of weld pool surface could be extracted
from the reflected points in captured images after image processing and
reconstruction algorithm was applied [23]. The welding torch can be controlled by the
Search WWH ::
Custom Search